[ 76/82] target/file: Use O_DSYNC by default for FILEIO backends

From: Ben Hutchings
Date: Fri Jun 08 2012 - 01:13:48 EST


3.2-stable review patch. If anyone has any objections, please let me know.

------------------

From: Nicholas Bellinger <nab@xxxxxxxxxxxxxxx>

commit a4dff3043c231d57f982af635c9d2192ee40e5ae upstream.

Convert to use O_DSYNC for all cases at FILEIO backend creation time to
avoid the extra syncing of pure timestamp updates with legacy O_SYNC during
default operation as recommended by hch. Continue to do this independently of
Write Cache Enable (WCE) bit, as WCE=0 is currently the default for all backend
devices and enabled by user on per device basis via attrib/emulate_write_cache.

This patch drops the now unnecessary fd_buffered_io= token usage that was
originally signalling when to explictly disable O_SYNC at backend creation
time for buffered I/O operation. This can end up being dangerous for a number
of reasons during physical node failure, so go ahead and drop this option
for now when O_DSYNC is used as the default.

Also allow explict FUA WRITEs -> vfs_fsync_range() call to function in
fd_execute_cmd() independently of WCE bit setting.

Reported-by: Christoph Hellwig <hch@xxxxxx>
Cc: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
Signed-off-by: Nicholas Bellinger <nab@xxxxxxxxxxxxxxx>
Signed-off-by: Ben Hutchings <ben@xxxxxxxxxxxxxxx>
[bwh: Backported to 3.2:
- We have fd_do_task() and not fd_execute_cmd()
- Various fields are in struct se_task rather than struct se_cmd
- fd_create_virtdevice() flags initialisation hasn't been cleaned up]
---
--- a/drivers/target/target_core_file.c
+++ b/drivers/target/target_core_file.c
@@ -134,21 +134,11 @@
ret = PTR_ERR(dev_p);
goto fail;
}
-#if 0
- if (di->no_create_file)
- flags = O_RDWR | O_LARGEFILE;
- else
- flags = O_RDWR | O_CREAT | O_LARGEFILE;
-#else
- flags = O_RDWR | O_CREAT | O_LARGEFILE;
-#endif
-/* flags |= O_DIRECT; */
/*
- * If fd_buffered_io=1 has not been set explicitly (the default),
- * use O_SYNC to force FILEIO writes to disk.
+ * Use O_DSYNC by default instead of O_SYNC to forgo syncing
+ * of pure timestamp updates.
*/
- if (!(fd_dev->fbd_flags & FDBD_USE_BUFFERED_IO))
- flags |= O_SYNC;
+ flags = O_RDWR | O_CREAT | O_LARGEFILE | O_DSYNC;

file = filp_open(dev_p, flags, 0600);
if (IS_ERR(file)) {
@@ -400,26 +390,6 @@
transport_complete_sync_cache(cmd, ret == 0);
}

-/*
- * WRITE Force Unit Access (FUA) emulation on a per struct se_task
- * LBA range basis..
- */
-static void fd_emulate_write_fua(struct se_cmd *cmd, struct se_task *task)
-{
- struct se_device *dev = cmd->se_dev;
- struct fd_dev *fd_dev = dev->dev_ptr;
- loff_t start = task->task_lba * dev->se_sub_dev->se_dev_attrib.block_size;
- loff_t end = start + task->task_size;
- int ret;
-
- pr_debug("FILEIO: FUA WRITE LBA: %llu, bytes: %u\n",
- task->task_lba, task->task_size);
-
- ret = vfs_fsync_range(fd_dev->fd_file, start, end, 1);
- if (ret != 0)
- pr_err("FILEIO: vfs_fsync_range() failed: %d\n", ret);
-}
-
static int fd_do_task(struct se_task *task)
{
struct se_cmd *cmd = task->task_se_cmd;
@@ -434,19 +404,21 @@
ret = fd_do_readv(task);
} else {
ret = fd_do_writev(task);
-
+ /*
+ * Perform implict vfs_fsync_range() for fd_do_writev() ops
+ * for SCSI WRITEs with Forced Unit Access (FUA) set.
+ * Allow this to happen independent of WCE=0 setting.
+ */
if (ret > 0 &&
- dev->se_sub_dev->se_dev_attrib.emulate_write_cache > 0 &&
dev->se_sub_dev->se_dev_attrib.emulate_fua_write > 0 &&
(cmd->se_cmd_flags & SCF_FUA)) {
- /*
- * We might need to be a bit smarter here
- * and return some sense data to let the initiator
- * know the FUA WRITE cache sync failed..?
- */
- fd_emulate_write_fua(cmd, task);
- }
+ struct fd_dev *fd_dev = dev->dev_ptr;
+ loff_t start = task->task_lba *
+ dev->se_sub_dev->se_dev_attrib.block_size;
+ loff_t end = start + task->task_size;

+ vfs_fsync_range(fd_dev->fd_file, start, end, 1);
+ }
}

if (ret < 0) {
@@ -478,7 +450,6 @@
static match_table_t tokens = {
{Opt_fd_dev_name, "fd_dev_name=%s"},
{Opt_fd_dev_size, "fd_dev_size=%s"},
- {Opt_fd_buffered_io, "fd_buffered_io=%d"},
{Opt_err, NULL}
};

@@ -490,7 +461,7 @@
struct fd_dev *fd_dev = se_dev->se_dev_su_ptr;
char *orig, *ptr, *arg_p, *opts;
substring_t args[MAX_OPT_ARGS];
- int ret = 0, arg, token;
+ int ret = 0, token;

opts = kstrdup(page, GFP_KERNEL);
if (!opts)
@@ -534,19 +505,6 @@
" bytes\n", fd_dev->fd_dev_size);
fd_dev->fbd_flags |= FBDF_HAS_SIZE;
break;
- case Opt_fd_buffered_io:
- match_int(args, &arg);
- if (arg != 1) {
- pr_err("bogus fd_buffered_io=%d value\n", arg);
- ret = -EINVAL;
- goto out;
- }
-
- pr_debug("FILEIO: Using buffered I/O"
- " operations for struct fd_dev\n");
-
- fd_dev->fbd_flags |= FDBD_USE_BUFFERED_IO;
- break;
default:
break;
}
@@ -578,10 +536,8 @@
ssize_t bl = 0;

bl = sprintf(b + bl, "TCM FILEIO ID: %u", fd_dev->fd_dev_id);
- bl += sprintf(b + bl, " File: %s Size: %llu Mode: %s\n",
- fd_dev->fd_dev_name, fd_dev->fd_dev_size,
- (fd_dev->fbd_flags & FDBD_USE_BUFFERED_IO) ?
- "Buffered" : "Synchronous");
+ bl += sprintf(b + bl, " File: %s Size: %llu Mode: O_DSYNC\n",
+ fd_dev->fd_dev_name, fd_dev->fd_dev_size);
return bl;
}

--- a/drivers/target/target_core_file.h
+++ b/drivers/target/target_core_file.h
@@ -18,7 +18,6 @@

#define FBDF_HAS_PATH 0x01
#define FBDF_HAS_SIZE 0x02
-#define FDBD_USE_BUFFERED_IO 0x04

struct fd_dev {
u32 fbd_flags;


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/